Agent Search MCP vs Web Retrieval MCP

free_search

Search the web with an explicit adapter set and bounded fallback. Best for: Quick fact-finding, general search, when date/domain filters are not needed. Not recommended for: Filtered or verified-only results — use free_search_advanced. For full page content — use free_extract. Twelve adapters are selectable; the default request uses DuckDuckGo + Sogou only. Adapters that share one upstream family are tried sequentially on failure and never double-count as corroboration. Explicitly requested optional API adapters run only when credentials are present and the free basket is short or below the quality gate. Results are deduplicated and include separate confidence, relevance, and source-count signals. @readOnly true @idempotent true — makes outbound HTTP requests to configured search engines. Injection detection and SSRF protection active.

free_search_advanced

Advanced search with filters and quality control. Best for: Domain filtering, high-confidence only, Chinese content. Not recommended for: Simple queries — use free_search instead. @readOnly true @idempotent true — runs waterfall progressive search across policy-allowed engines. Makes outbound HTTP requests to search engines and optionally to Jina Reader for content enrichment.

free_extract

Extract full content from a URL. Returns clean markdown text. Best for: Reading a specific page found in search results to get full context. Not recommended for: Bulk extraction — use free_search first to find relevant pages. Behavior: Makes an outbound HTTP request to Jina Reader (r.jina.ai) which fetches and converts the page to markdown. Has SSRF protection: blocks private IPs, localhost, and metadata endpoints. 10s request timeout — pages exceeding this will fail with a timeout error. HTTP errors (4xx, 5xx) are returned as structured error responses.